China 
Africa 
Explore chapters and articles related to this topic
Some (Not So) Boring Stuff
Published in James Luke, David Porter, Padmanabhan Santhanam, Beyond Algorithms, 2022
James Luke, David Porter, Padmanabhan Santhanam
Like the DevOps process for software code and its configurations, MLOps represents the step of moving the validated AI model from development to operational use and managing it in a production environment. We refer to [33,34] for detailed descriptions and underlying practices. MLOps is a complicated extension to DevOps in that the ML models are easily changed, yet have a significant impact on business decisions and must therefore be carefully managed.
What drives MLOps adoption? An analysis using the TOE framework
Published in Journal of Decision Systems, 2023
Sibanjan Debeeprasad Das, Pradip Kumar Bala
The heavy lifting of unifying the ML Development process and ML System Operations is called MLOps (Google Cloud, 2020). It is a set of technology, practices and processes that aims to provide a system in place in order for organisations to develop AI/ML models, implement and monitor the ML workflows. Just like IT operations is required to ensure an organisation’s technology systems run smoothly to facilitate the business functions achieving their goals, MLOps is necessary for driving effective and efficient use of AI/ML models used by the business functions to drive their objectives. MLOps is eventually an essential component in the ML ecosystem that brings the ML models’ data pipeline, model pipeline and release pipeline together. The primary goal of MLOps is to maximise ML model performance and increase agility in model development, release and post-production process to ensure the right results are available to authorised users at the right time (Subramanya et al., 2022). It involves automating various components of an ML system, streamlining the model development life-cycle, placing controls for better model governance and providing a platform for model serving and continuous monitoring for risk management. As many functions come together to create an MLOps system, it takes much effort and commitment to implement and adopt this technology stack. Several factors drive the usage of MLOps tools, such as improving the availability of ML services, capturing user feedback, improving data quality or improving the AI-ML team’s experience in automating some of their day-to-day operational tasks (Zhou et al., 2020). Some factors might impede people from using the system, such as the availability of MLOps experts or tools in the team, significant investment costs or low tolerance to experiments on new technologies. These factors form the barriers to adopting MLOps technologies.